The AI's greatest flaw and improvement suggestion Thread - Conquest of Elysium 3

I am playing a Troll game on Dark Ages vs. 3 Knight AIs (at least I think so, started yesterday and forget how to see settings in-game).
Two AIs have already died, it is Askuth the cultist left with 5 recruitment centers (mostly old castles). I have four dragons for scouting and resource stealing, and quite a few quasi-solo goblin scouts hiding in forests for recon, and I am shocked at what he is spending his iron and money on: he has huge numbers of catapults and heavy infranty (alright) in his various citadels, but all of these are without commanders. He has only two commanders left now, one with a decent force of 90 units, and one with 30, while he has about 400 units (from what my recon can gather) in citadels without commanders. After I killed his starting High Cultist, I have been scouting the last couple of turns, and the AI has NOT purchased any additional commanders, but he does seem to spend his money on fodder.

I am convinced that the AI does not need more cheats to become more challenging; it has, in this game, money in abundance. But for some reason, it does not choose to get commanders often enough. So far, this has always been his downfall. I am watching him gather more and more troops, and I think if he had thrice the resources he would still be functionally commanderless. This really hampers the AI. It drastically lowers its mobility, and hence its ability to perform resource-grabs, get troops to the front, do recon, or make attacks that are not incredibly easy to spot.

The AI seems interested enough in resources. And it will often not walk straight into its own doom. This is good.

But I would suggest changing the buying habits of all of the AIs I have seen so far (I've seen everyone but the dwarf queen and the barbarian as AI) to emphasize commanders over troops, even more so if commanders are less than 4 in number, much more so if less than 3, with highest priority if less than 2, more so again with nations with free-spawn, and more so again with nations with access to sufficient fodder-troops as summons.

Make the AI prioritize commanders after the early game and its performance will increase dramatically.

Valid points. I would also like to see the AI be far more aggressive against citadels defended by a player. I have now been many times bypassed entirely by AI forces that would absolutely flatten me, whether I had a defense bonus of 1, 2 or 3. If I have a bonus of 2 or 3, I won't get attacked. If I have a bonus of 1, I may be attacked.

This drags games on and on even if I have no realistic chance to do anything. Only if I come out from behind the walls will I be attacked. So the AI targeting and risk assessment algorithms need some serious work with regard to how it deals with fortified positions like citadels.

At the moment I am usually playing against Baron AI and I have to agree with the OP. I would also suggest making more commanders available for the AI to hire with increased difficulty level (handing them more cheats).

There are also additional pressing issues with the AI: Many times they die within the first turns of a game - a workaround I could think of would be giving each class an immobile commander unit at their starting locations that has some defensive capabilities (similar to the barbarian totem - just make it a commander).
If my first suggestion (more available commanders for the AI) gets integrated and considering the fact that they get additional income with increased difficulty level, they will hopefully be able to recover from a loss of all their mobile commanders.

Then there's the issue of army management: I have yet to fight a lategame battle that would pose a significant challenge to my main army. The problem seems to be that the AI prefers to send its troops piece-meal, around 50 units at a time. Rarely have I seen 100 units or more and even those armies are barely any threat: Because my main-army will have powerful mage support (having several lvl 3 mages and an abundance of lower lvl mages together with heavy tanks beats almost any army in the game), while the AI seems to be reluctant to have more than 1 commander per army.
This makes lategame feel like fighting a guerilla-war, not a series of epic battles for strategic key locations (I would love to actually have to lay siege to the enemy capital).
Sending out smaller contingencies of troops is fine - but round-about 10 soldiers with a cheap commander should be enough for site- or forest-grabbing, whereas 30 or 50 soldiers are too costly for that endeavour. Thus lategame, I would suggest tweaking how the AI builds armies: There need to be one or few main-armies with several mages and high-tier troops that will try to grab and hold key-locations of the map (your capital, their capital, a gold mine, a densely populated area - that kind of stuff) and multiple small, preferably mobile, troops that try to steal your resources and flee from bigger forces. Ideally, of course, the AI would have to assess the situation by measuring how many troops you use to defend sites and how big your biggest armies are and adjust - always having a few troops more than you for the appropriate task. But I assume that would be hard to code.

I think in general the AI plays too aggressively. Throughout the game, I have often seen it suicide army after army against me until it ran out of steam and just died without me having to actually invade. (Note that it rarely actually engages in combat - it will just send medium-sized armies to try and grab sites that are scattered around key-locations. These key-locations, however, are defended by my main army. The AI seems to be oblivious to that fact.) Ideally I would like to be able to set the AI's aggression from very aggressive to very defensive in addition to the difficulty-level-setting.

Edirr wrote: Valid points. I would also like to see the AI be far more aggressive against citadels defended by a player. I have now been many times bypassed entirely by AI forces that would absolutely flatten me, whether I had a defense bonus of 1, 2 or 3. If I have a bonus of 2 or 3, I won't get attacked. If I have a bonus of 1, I may be attacked.

This drags games on and on even if I have no realistic chance to do anything. Only if I come out from behind the walls will I be attacked. So the AI targeting and risk assessment algorithms need some serious work with regard to how it deals with fortified positions like citadels.

Yeah, the AI needs to learn how to lose some battles in order to win the war as well. I've seen it bypass troops of mine that it could have crippled and then killed with another army and make overall poor use of its numerical superiority. When the AI has an abundance of troops, it seems to me that it needs to be more willing to take risks. I've seen Hoburg armies of 200 bypass my own of 50 when not 4 tiles away there were two more armies of 150 and 100 that it had in reserve.

I agree that the AI plays too aggressively with piecemeal armies.

Also when the AI has multiple stacks in range of an army of mine it doesn't seem to realise that if they all attacked together they would win.

Agreed with everything above.

Pros and Cons here.
With Dom3 we could make general statements. But with CoE3 we have separate AIs.

What AI nation, did what against what human nation? And what specific recommendation would you make for that nations AI?

I've played quite a few games and reached the same conclusion. Upon some particular instances to where I *should* have been completely doomed and wiped out in one turn by a multiple flank attack, I consistently watched the AI spread out to reclaim the lost resources I just took instead. There doesn't appear to be a collective unit tally check to initiate a "Stack of Death" attack which the human player strives for ultimately. Consider it a gift horse or perhaps a feature.

Hmm... so its a general tactic that should be applied to all of the AIs?
Well not ALL of the AIs. Not the Independents. And not sure if it should be for team members.

All AI nations I've played against like to split their armies just in a wrong moment.
The AI comes with an impressive horde, you retreat, it capture some mine and leaves part of the army to guard it. If the AI has extra commanders it can immediately split the rest of the army in two, which then go to grab farms or something. Suddenly, you face not a mighty horde but a three relatively small armies you can easily defeat in detail.
Now, splitting armies is not a bad thing. If the AI always concentrated all its forces in one single army that too would be easily exploitable. But spreading out when there is a strong enemy army withing a striking distance is a bad idea.
Not exactly the same problem, but somewhat related: the Dwarf AI establishes a new colony on the border. My strong army stands next to that mine. Next turn he goes away, leaving only a Dvala with a single guard in a new colony. Sure, the Dvala can't leave. But why to establish a colony in the first place? My army already was there.

Gandalf, I tried in my original post to indicate that these are changes that need to be made to all 17 class AIs, with some differentiation as follows: classes with access to freespawn fodder (e.g. Bakemon, Cultist) and to a just slightly smaller degree commander-fed fodder (e.g. Priest King, Necro, etc.) need even more commander purchase priority. That the AI should, if it cannot do a troop- or army-assessment check, simply count the years (after 3 years, increase commander priority, after 5 years, significantly increase it, after 7, drastically increase it), but ideally, the AI should assess its army strength (infantry count, archer count, perhaps HP count, damage count) and if the ration of commanders to non-commander units is less than, say, 1/X (where X might start at 40 as a default for many nations but get smaller as the years go on, perhaps smaller for nations such as the Troll, ) then it should NOT buy troops, but rather horde its money for when commanders appear. I also suggested elsewhere (in the Features request thread) that starting with Knight and above, the AI might have an increased chance of commander special recruitment (to facilitate its ability to spend its money wisely).

I do not see the "pros and cons" you mention, unless by "con" you mean time spent on updating the algorithms. That's for sure.

Admittedly, I am practically computer illiterate and cannot program AI.

The lack of aggressiveness in taking occupied citadels is more or less a problem for all AIs. Or any structures that give defense bonuses, for that matter. The specifics per AI may vary somewhat, but more aggression when they have a significant numerical superiority (weighted by special abilities) would need to be added.

I play on a small map with two random opponents and here is what I am seeing. The AI wanders into my area taking resources usually with 2 smaller stacks. it does not realize I have a stronger army and I quickly dispatch it. This happens almost every game. I get the sense there is no AI - it seems it is wandering around grabbing resources. I get no sense it is assessing the situation and trying to make a good decision.

Maybe if I played with one opponent or a bigger map I would see bigger stacks on my area. But, I just don't want big stacks being streamed at me. I want an Ai that is attempting to play the game semi-intelligently.

Considering there is no async multiplayer this is a huge issue.

I also agree regarding the AI's avoidance of defensive structures and lack of aggressiveness therein, as well as its inability to assess the map.

Is there any chance, Edi, that you could make a very small section "AI" in the distilled Wish Shortlist Thread and include a thing or two here?

What we really need is moddable AI. Once we have it, we can stop whining about the AI being stupid and start writing our own stupid AI.

The AI can group units into a large army (in fact on the Indepts Lord scenario map its quite an irritant).
And if it finds a place that it cannot take, it does seem to hang around the area with that army while starting another army toward it.
I have seen it attack with two armies at once but I think that might just be timing. They both got there at the same time.

So what seems to be lacking from what a Human would do, is that the AI doesnt put two commanders with armies together into one army, in order to take a hard spot. and then break them back up again so he isnt always marching one giant army around.

I suppose the breaking back up isnt hard. It will just consider everything to be like a recruit center even if it isnt. That would have it releveling the units between the commanders (and hopefully grouping by movement abilities). A nice bonus to another thing it needs to do more often.

The grouping might be more involved. What would be the formula? "how many commanders do I have within X moves of the target". Reassign those commanders to move to and join with the commander already waiting there.

But if it does this every time it finds a target it cant immediately take, would that kill its expansion? And how much "too much to take" should it tackle? Players only? Resources? If it does that for everything it finds then it will send everything it has to Gold Mine or a beholder held location. The game does have army-strength numbers it uses. So maybe "if enemy army greater than my army but NOT greater than 1/2 of all of my armies"?
(I know this isnt a question for everyone here but there are some here who might tackle it)

Edited by: gp1628

Another thing the AI seems to be weak at is complementing an army with spellcasters. It looks like the AI thinks all commanders are there to lead armies and so it do not employ them for anything else, be it recon or casting spells. Say, the Troll King AI doesn't seem to make a good use of different shamans available to it. Having 3-4 shamans/witch doctors behind rock trolls can make quite a difference in combat... but I do not see the AI using them that way.

Good point. That would be one-leader = one-army. Which is probably ok for expansion.

But this "group commanders for a big strike" would tend to fix that. For larger, more difficult targets, it would be commanders together for one attack so there would be mages fighting with captains.

Its a tricky thing to do an AI for this game as ideally you want to be able to defeat the enemy's largest army and with what ever excess run into his backfield and claim resources to defeat his economy. Currently the AI seems to largely ignore building an army to defeat the players largest army and just tries to claim resources. I have seen large armies in game but thus far I've had a stronger army at the time to defeat it.

Oh, I have lost the game but that's at higher difficulties where the standard AI army swarm wears me down faster, or their early game is better than mine and I lose to their first army.

Edited by: ZephyrDoo

I like playing on tiny or small maps with one opponent for quicker games and here's what I am seeing:
- In just about every game it places my opponent right next to me. I suggest having a minimum distance between opponents and not make it too small.
- The AI, since it is almost always next to me, trickles in a few small stacks, which I quickly dispatch and I win since one stack has their leader. It has no sense it is in danger, it
just blindly goes after resources. How about some intelligence when it is in danger or in enemy territory?

please make ai moddable if you can

Some observations from my current game:

It seems that the Necromancer AI overuses Raise Dead. In this game I'm playing against 3 allied AI, so the only real opposition the Necromancer faces is me. It's Empire society, so there should be no very strong indies. Still all his commanders I've met so far had some 50 points of Insanity... and not that many undead with them.

A Witch with a small army containing a Creeping Doom happily attacked my castle guarded by a tiny garrison and a couple of trebuchets. Sure, she took the castle. Just as sure, she died to the thebuchet fire and it looks like she was the only fungi user the AI had in the area.
From what Gandalf said I understand that the AI bases its attack decisions on the estimates of opposing armies strengths. Maybe it would be better if it took a Monte Carlo approach, playing out the battle many times to see which outcomes are likely. A likely loss of a key commander for a very little gain should not be considered an acceptable outcome.

Edited by: jsv

Also some observations:

1. Witches doesn't seem to actively capture forest and jungles (maybe swamp too). I played a game allied with a witch AI, and it seems that after the 1 year they just gave up on forest/jungles, they went after settlements and resource nodes just fine, but completely ignored the swath of forest within easy reach, any forest they take are only if its part of the pathing, this looks permanent (I guess Im 6-7 years in?). I presume the same AI flaw applies to druid and barbarian too.

2. Enchanter AI has a fetish about terracotta soldiers, which is a deathtrap. In fact they are so addicted that once they find a swamp (they'll make a beeline for it), they won't move away from it until its depleted. I was able to past right by him with a far weaker army and take over all his stuff with no reaction. After he depleted the swamp he moved to another swamp that's really close without retaking his stuff. I laughed, took his last fortress, and he died with a 80+ strong army still raising terracottas.

Those seem like bugs to me, in the sense that the AI should devote a significant amount of effort to capturing special resource generators.

I've added the issues about the AI to the shortlist.

I have an impression that the AI puts very little priority on expansion and a huge priority on retaking what it had lost. It often sits in relatively small area, locked in a fight with a neighbor over a couple of farms or hunting deer, while ignoring vast unclaimed lands around.

Edited by: jsv